Governance needs a memory.

Context engineering

Context engineering is the practice of giving an AI system the right information at the right time: the system prompt, the tools it can call, examples of the behavior you want, the running message history, what it retrieves, and what it writes down for later. Memolok is the Decision Record: what your organization decided, captured as you work.

Memolok is the governance layer

Every other piece of an agent's context, the system prompt, its tools, what it retrieves, the running message history, is built to be read and replaced. Governance is different: it's where an agent's rules come from, not one more thing it reads. Memolok is that piece, the sealed, permanent account a rule can be checked against, kept apart from the conversation that produced it.

Where the Decision Record fits

Anthropic's own description of context engineering names the same few moving parts most practitioners now use: the system prompt, the tools an agent can call, examples of good behavior, the running message history, what gets retrieved on demand, and what gets written down and carried forward for anything that runs longer than one session.

A Memolok Decision Record doesn't live in any of those slots. Once it's approved, it doesn't get overwritten, compacted, or silently revised: what was decided, what else was considered, and what was expected to happen stays exactly as recorded, kept apart from the conversation that produced it.

What a policy trace tells you, and what it doesn't

Some agent systems log what a rule did: an exception granted, a threshold crossed, an override applied. That's proof the rule executed, not proof anyone ever decided it, or why. It stays invisible until someone finally asks why the threshold is what it is, and there's no answer waiting.

That's the gap a Decision Record closes. The policy itself, why this threshold, what alternatives were considered, what was expected to happen, is an entirely ordinary Memolok Decision Record: human-made, accountable, sealed once it's settled. The trace tells you the policy ran. The record tells you why it exists.

Governance at any scale

Governance sounds like an enterprise word, but the same shape holds at any scale: a solo builder deciding how their own project should behave, a team setting a policy for how they work, or a whole organization governing what an AI agent is allowed to do on its behalf. See the Use cases page for what that looks like at each size.

Try Memolok today

Memolok is in early testing, and we're inviting people to try it now. Sign up today and start recording your decisions. A person reads that list, so nothing lands in your inbox automatically.

Common questions

Where does a Decision Record fit in context engineering?

Alongside the other things an agent's context holds: its instructions, its tools, what it retrieves, what's in the conversation so far. Most of those are built to be read and replaced. A Decision Record is different: a governance concept, not another piece of working memory. Once a decision is approved, it doesn't get silently rewritten, only superseded, so an agent can always tell what's still in force.

Isn't a trace of what an agent did the same thing as a decision record?

No. A trace shows a policy running: an exception granted, a threshold crossed. It's evidence the policy executed, not evidence anyone ever decided it, or why. A Memolok Decision Record is the policy-setting decision itself: the alternatives considered, what was expected, and who committed to it.